Re: gnulib-tool.py: Optimize --extract-recursive-dependencies.

[Bruno Haible]

Hi Collin,

> A while ago I said that I would have a look at optimizing the
> --extract-recursive-dependencies option you implemented in gnulib-tool.
> Then I forgot about it.
> 
> I've written and pushed the change finally.
> 
> Here was my testing:
> 
>     $ time gnulib-tool --extract-recursive-dependents c99 > 
> extract-recursive-dependents-c99-old
> 
>     real      3m53.437s
>     user      0m41.646s
>     sys       3m24.750s
>     $ time gnulib-tool --extract-recursive-dependents c99 > 
> extract-recursive-dependents-c99-new
> 
>     real      0m7.485s
>     user      0m7.078s
>     sys       0m0.312s
>     $ cmp extract-recursive-dependents-c99-old 
> extract-recursive-dependents-c99-new 
>     $ echo $?
>     0

Nice. Very nice! Thank you.

Could you please fix the ChangeLog entry (s/dependencies/dependents/)?

And I also see a performance regression in one case:

$ time ./gnulib-tool --extract-dependents stdlib
$ time ./gnulib-tool --extract-dependents stdlib
$ time ./gnulib-tool --extract-dependents stdlib
Before: ca. 0.2 to 0.25 seconds.
AFter:  1.0 seconds.

In this case, when there is only a single call to getDependents(), it was
apparently faster to use the 'find modules ...' approach, compared to
_getAllModules().

Could you please make this case fast again? The way I'm thinking of is to add
a boolean argument 'top_level_call' to getDependents(), and pass this argument
as true from main.py and as false from everywhere else.

Bruno